ModelBrief · Qwen4-Max-Preview · Topic Page
Last updated Nov 13, 2026

A Third-Party Per-Token Price Appears: $0.353/$1.059 per 1M — Plus a 300-Token Empty Reply

Published Nov 13, 2026 · Last updated Nov 13, 2026

Alibaba published no per-token price for Qwen4-Max-Preview; the only official route is a Token Plan subscription. On November 11, third-party platform AIHubMix became the first to list the model pay-per-token — and, in passing, caught an early bug worth documenting. Here are the two findings, kept carefully separate.

The price: $0.353 in / $1.059 out per million tokens

AIHubMix lists input at $0.353 and output at $1.059 per million tokens. Emphasis first: this is a third-party platform's own pricing, not Alibaba's. It does not represent the real cost of Token Plan access and belongs in no "official price" comparison. Its value is as a market anchor — the number future official pay-per-token pricing will be compared against.

The bug: 300 tokens billed, nothing returned

AIHubMix observed a reply where roughly 300 tokens were generated on the meter yet the returned content was empty. Day-one preview bugs like this are unremarkable — streaming, filtering or metering are all candidate culprits, and nothing is confirmed. The practical lesson for developers: any integration (official terms already bar production automation) must handle empty replies with retries and validation.

Relation to official channels

For the record: the official route remains Token Plan only, from $6/month (new users $3 for the first month), restricted to personal interactive development with no production automation or batch scripting. The compliance and stability of third-party channels are outside this article's scope.

Our take: a short post with double value — the market's first price reference, and a reminder that a preview is a preview: metering and output can both glitch. If you cite the price, cite it as third-party pricing, every time.

Reference source: this article is an original rewrite produced by this site; facts and figures were checked against AIHubMix(Medium)(2026-11-11)

← Back to the Qwen4-Max topic page